Agentic Engineering Patterns (simonwillison.net)

Article summary

The article discusses agentic engineering patterns, which are principles for effectively using coding agents like Claude Code and OpenAI Codex to produce better code. It emphasizes the importance of having good tests, avoiding technical debt, and using coding agents to handle repetitive tasks. The article also explores the concept of the compound engineering loop, where coding agents and humans work together to improve code quality. By following these patterns, developers can leverage the strengths of coding agents to improve their workflow.

Main themes

• Agentic Engineering
• Coding Agents
• Software Development
• Testing and QA
• Technical Debt
• Collaborative Coding

What commenters say

• Some commenters find that coding agents are not yet reliable for complex tasks and can produce low-quality code, while others have had success with them for certain types of projects.
• The effectiveness of coding agents depends on the quality of the tests and the ability to validate their output, with some arguing that they are only useful for trivial boilerplate code.
• Others believe that coding agents can be useful for tasks such as navigating unfamiliar codebases, but may struggle with more complex calculations or logic errors.
• There is disagreement about the rate of progress in coding agent technology, with some arguing that significant improvements have been made in recent months, while others are skeptical and believe that the technology is still evolving too rapidly.
• Some developers have found that using coding agents can save time and improve productivity, especially when used in conjunction with planning and verification steps.
• Others argue that relying too heavily on coding agents can lead to technical debt and decreased code quality, and that human engineers need to maintain their skills to ensure responsible code ownership.